Annals of Medicine.

2009; 41: 522532

ORIGINAL ARTICLE

Haptoglobin and risk of myocardial infarction, stroke, and congestive heart failure in 342,125 men and women in the Apolipoprotein MOrtality RISk study (AMORIS)

INGAR HOLME1, ARE H. AASTVEIT2, NIKLAS HAMMAR3, INGMAR JUNGNER4 & RAN WALLDIUS5 GO
Department of Preventive Cardiology, Centre of Preventive Medicine, Oslo University Hospital, Ulleval, Oslo, Norway, Department of Chemistry, Biotechnology and Food Science, Norwegian University of Life Sciences, Aas, Norway, 3 Department of Epidemiology, Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden, and AstraZeneca R&D, So derta lje, Sweden, 4Department of Medicine, Clinical Epidemiological Unit, Karolinska Institutet, and CALAB Research, Stockholm, Sweden, and 5King Gustaf V Research Institute, Karoliniska Institutet, Stockholm and AstraZeneca, Sverige, So derta lje, Sweden
2 1

Abstract Background. Few studies have analysed the importance of haptoglobin (Hp) as a risk factor for acute myocardial infarction (AMI), stroke, and heart failure (HF) in large healthy populations. Aims. We examined Hp as risk factor for AMI, stroke, and HF in the Apolipoprotein MOrtality RISk (AMORIS) Study and compared its predictive strength with that of total serum cholesterol (TC). Methods. Prospective study (11.8 years) of AMI, stroke, and HF through linkage with Swedish hospital discharge and mortality registers with measurements of Hp in 342,125 subjects. Results. Hp is a significant risk factor of AMI, stroke, and HF. Relationships were stronger for men than women with regard to stroke and HF, but not AMI. Hp was almost as predictive as TC for AMI and about equally predictive of stroke with a stronger relationship to ischaemic than haemorrhagic stroke. A 4.2-fold increase in risk of AMI was observed comparing subjects in the joint lower quartiles of TC and Hp to those in the upper. For stroke, the risk increase was 2-fold, and 1.5-fold for HF. Interpretation. The inflammation marker Hp carried as much additional predictive information value on AMI and stroke as TC. Hp was also a risk factor of HF. Key words: Apolipoprotein, cholesterol, haptoglobin, heart failure, myocardial infarction, stroke, triglycerides

Introduction Inflammation has a central role in the development of atherosclerosis and premature cardiovascular morbidity and mortality (15). Many biochemical markers have been introduced to measure such inflammation, among which high sensitive C-reactive protein (hsCRP) is the most often used (6,7). Other inflammation-specific plasma proteins (ISP) are fibrinogen, ceruloplasmin, orosomucoid, a-1-antitrypsin, and haptoglobin (Hp) (35). When determining

their predictive power, these are often combined into a one-dimensional categorical variable, such as number of markers in the upper quartile of their distribution, so that cardiovascular risk assessments can be enhanced as compared to each single marker on its own (35). However, it is still unclear if such a combination is superior to each of the single markers since a test of the addition of the combined markers to each of its components on risk has to our knowledge not been reported. Hp may exist in three functional subtypes, all of which can express various degrees of

Correspondence: Ingar Holme PhD, Centre of Preventive Medicine, Oslo University Hospital, N-0407 Oslo, Norway. Fax: 47-22 11 99 75. E-mail: ingar.holme@uus.no (Received 11 March 2009; accepted 1 June 2009) ISSN 0785-3890 print/ISSN 1365-2060 online # 2009 Informa UK Ltd. DOI: 10.1080/07853890903089453

Haptoglobin and risk of AMI, stroke, and HF Key messages . Haptoglobin (Hp) was a significant risk factor for acute myocardial infarction (AMI), stroke, and heart failure. A 4.2-fold increase in AMI risk was observed comparing the joint lower quartiles of Hp and total serum cholesterol with the upper. Abbreviations

523

anti-inflammatory, antioxidant capacity. Hp inhibits prostaglandin synthesis and has angiogenic effects besides binding of haemoglobin and inhibiting toxic reactions from bacteria and viruses (8). Hypercholesterolaemia has been reported to be associated to ISPs (4). Also, total cholesterol (TC) seems to work interactively with ISPs on cardiovascular end-points, but formal tests of interactions were not reported (4). In the Apolipoprotein MOrtality RISk (AMORIS) study 342,125 healthy men and women in the age group 3085 years had measurements of Hp, an acute reactant inflammation plasma protein marker, and also measurements of lipids. These were followed with respect to development of cardiovascular disease end-points during 11.8 years. The aims of this study were to report with AMORIS data, by far the largest database available, on the relationship between Hp and cardiovascular end-points such as acute myocardial infarction (AMI), ischaemic stroke (IS), haemorrhagic stroke (HS), and congestive heart failure (HF) with or without a cardiovascular disease origin. We also wanted to explore the interactive nature of TC, triglycerides (TG), and Hp on such end-points showing the additional predictive information of Hp over and above that of TC and TG.

AMI acute myocardial infarction AMORIS Apolipoprotein MOrtality RISk study AP attributable proportion of risk due to interaction apoA-1 apolipoprotein A-1 apoB apolipoprotein B DM diabetes mellitus HF congestive heart failure Hp haptoglobin HR hazard ratio HS haemorrhagic stroke hsCRP high sensitive C-reactive protein ICD International Classification of Diseases IS ischaemic stroke ISP inflammation-specific plasma protein RERI relative excess risk due to interaction S synergistic index of Rothman SD standard deviation TC total serum cholesterol TG triglycerides

Patients and methods The AMORIS study population has been described earlier (913). In short, AMORIS included subjects who submitted a blood sample from health checkups and out-patient clinics mainly from the greater Stockholm area. No samples came from hospitalized patients. The inclusion period lasted from 1985 to 1996. Incident cases of type-specific strokes, acute myocardial infarctions, and heart failures in the AMORIS cohort were ascertained by record linkage to the Swedish national hospital discharge register and the Swedish national cause-of-death register. The hospital discharge register covers discharges with International Classification of Diseases (ICD)coded diagnoses from all Swedish hospitals treating acute medical cases regionally since 1964 and

nationally since 1987. Excluded were cases with non-fatal stroke (NFS), AMI, or HF as primary or secondary diagnosis prior to AMORIS blood sampling. The period of follow-up ended on 31 December 2002, and the average follow-up time was 11.8 (range 717) years. From the hospital discharge register and the cause-of-death register, respectively, all hospital discharges and deaths with stroke (ICD7: 330334; ICD-8 or ICD-9: 430438; ICD-10: I60I69) as diagnosis for subjects of the study population were extracted. In the analyses ischaemic/infarct stroke (IS) was defined as ICD-7: 332; ICD-8: 432434; ICD-9: 433434; and ICD-10: I63. HS was defined as ICD-7: 331; ICD-8 and ICD-9: 431; and ICD-10: I61. Diagnoses for HF were ICD-7: 422, 434.1, 434.2, 434.4, and 782.4; ICD-8: 427.0, 428, 429, 782.4; ICD-9: 428; ICD10: I50. Likewise, all hospital discharges and deaths with AMI (ICD-8 and ICD-9: 410 and ICD-10: I21) as diagnosis for subjects of the study population were extracted. The validity of the hospital discharge and mortality data in Sweden concerning the AMI and stroke diagnoses has been evaluated repeatedly and found to be reasonably high for epidemiological purposes, as referred by us previously (1420). The validity of HF has been investigated in administrative registers in Denmark and Sweden and was found to be very specific, but a clear under-reporting was indicated in the Danish study (21,22). TC and TG were measured enzymatically as described previously (13,23,24). Fasting state was reported for 62% in AMORIS, and analyses

524

I. Holme et al. quartiles, and the lowest combined category was used as reference in a categorical Cox regression model estimating hazard ratios of AMI, IS, and HF in the 15 combined categories compared to the reference, adjusting for age, gender, TG, hospitaldiagnosed hypertension, and diabetes mellitus (DM). Graphs displaying the variation in relative risks by these combined categories were then created. Interaction tests were performed between TC, TG, and Hp on AMI, IS, and HF by creating a product term between TC or TG and Hp in addition to each of TC, or TG and Hp and adjustment factors in the Cox model, and its significance was tested by a Wald test. In addition we calculated three indices associated to synergistic or antagonistic effects of Hp and TC on end-points (25,26) by using logistic regression on each end-point as dependent variable and Hp and TC below or above their median as exposure variables, adjusted as above. One index estimates the relative excess risk due to interaction (RERI). Another is the attributable proportion due to interaction (AP). A third is the Rothman index (S) of synergism calculating the ratio of the joint effect of Hp and TC to the theoretical joint effects on the assumption of independence between them (27). S greater than 1.0 indicates synergism, that is an effect larger than the added effects of the Hp and TC, and S B1.0 indicates antagonism. Since occurrence of an AMI prior to a HF could influence the subsequent risk of HF, a special timedependent covariate Cox regression analysis was run treating occurrence of an AMI prior to the first occurrence of HF minus 90 days as a dichotomous additional adjustment variable. The 90 days lag time was chosen rather arbitrarily to avoid an AMI occurrence too close to the first HF. In addition a special Cox regression analysis was performed in the subgroup that had an AMI prior to a discharge from hospital with HF.

involving TG were adjusted for fasting state. Analyses were also done when only subjects with known fasting status were included. The results were practically the same and are not given. Hp was measured by an immunoturbidimetric method (reagents from Orion Diagnostics, Finland). The total imprecision calculated by the coefficient of variation was 5.6% at Hp level 1.1 g/L. In a subgroup of subjects, apolipoprotein B (apoB) and apolipoprotein A-1 (apoA-1) were also measured by immunoturbidimetric methods (23,24). All methods were fully automated with automatic calibration and laboratory facilities accredited (24). Hypertension status was not recorded in these subjects. However, 1,966 (0.6%) subjects hospitalized at any time prior to blood sampling with a primary or secondary hospital discharge diagnosis of hypertension were classified as hospital-recorded hypertensive subjects (ICD-8 codes: 400404; ICD-9 codes: 401405; ICD-10 codes: I10I15), although it is realized that such a method is conservative as compared to a direct method by blood pressure measurements or by questionnaire. Diabetes was considered present in those with fasting glucose measurement 7.0 mmol/L at baseline or with a discharge diagnosis of diabetes (ICD7: 260; ICD-8: 250.0259.9; ICD-9: 250; ICD-10: E10E14) prior to blood sampling. There were 10,037 (3.0%) classified as diabetics. Analyses in this report include all AMORIS persons 3085 years of age, free of registered stroke, AMI, or HF at blood sampling that included valid measurements of TC, TG, and Hp, totalling 342,125 subjects of whom 184,261 were men and 157,864 women. Statistical methods Differences of biochemical variables by quartiles of Hp were evaluated by analysis of variance with linear contrasts for continuous variables and chi-square for tests of trend for categorical variables. We analysed the relationship between time from blood sampling to first occurrence of cardiovascular events or end of follow-up and levels of Hp by Cox regression analysis. For analysis of total stroke all deaths except stroke deaths were censored; similar analyses were conducted for ischaemic and haemorrhagic strokes, AMI, and HF. Hazard ratios by gender were expressed by 1 standard deviation (SD) difference in the variables including 95% CI and were adjusted for age alone or age plus TC, TG, hospital-diagnosed hypertension, and diabetes status. Differences in hazard ratios were evaluated by Z-tests. Hp and TC were cross-classified by

Results During the study there were 8,463 men and 6,072 women who had a first stroke, 11,216 and 4,291 who had a first AMI, and 4,670 and 3,634 who had a first HF, respectively, as shown in Table I. Hp was associated with age but not with gender. TC and TG were positively and significantly (analysis of variance) but not strongly related to the level of Hp in both genders. The prevalences of hospital-diagnosed hypertension and DM were about doubled from quartile 1 to quartile 4 of Hp. However, the special recording of these two clinical variables gave gross under-estimates of their true prevalences, especially

Haptoglobin and risk of AMI, stroke, and HF

Table I. Base-line characteristics (mean (SD)) by gender and quartiles Female Hp categories No. of patients No. of strokes (%) No. of AMI (%) No. of HF (%) Age, years Haptoglobin g/L Total cholesterol mmol/L Triglycerides mmol/L Hypertension (%) DM (%) Male Hp categories No. of patients No. of strokes (%) No. of AMI (%) No. of HF (%) Age, years Haptoglobin g/L Total cholesterol mmol/L Triglycerides mmol/L Hypertension (%) DM (%) Q1 50.9 33,864 845 (2.5) 522 (1.5) 508 (1.5) 46.7 (11.4) 0.72 (0.10) 5.53 (1.12) 0.92 (0.51) 139 (0.4) 498 (1.5) 50.8 42,507 1,334 (3.1) 1,542 (3.6) 614 (1.4) 45.2 (10.4) 0.70 (0.11) 5.60 (1.05) 1.28 (0.75) 177 (0.4) 1,025 (2.4) Q2 0.91.0 44,771 1,378 (3.1) 807 (1.8) 710 (1.6) 47.8 (11.7) 0.95 (0.05) 5.62 (1.12) 1.00 (0.56) 207 (0.5) 735 (1.6) 0.81.0 47,154 1,673 (3.5) 2,157 (4.6) 902 (1.9) 46.6 (10.7) 0.95 (0.05) 5.69 (1.05) 1.37 (0.77) 201 (0.4) 1,349 (2.9)
a

525

of Hp in AMORIS population free of AMI, stroke and HF. Q3 1.01.2 40,985 1,669 (4.1) 1,074 (2.6) 928 (2.3) 49.4 (12.2) 1.14 (0.05) 5.77 (1.15) 1.13 (0.62) 232 (0.6) 880 (2.1) 1.01.3 45,054 2,186 (4.9) 2,852 (6.3) 1,077 (2.4) 47.9 (10.9) 1.14 (0.06) 5.80 (1.08) 1.47 (0.81) 255 (0.6) 1,654 (3.7) Q4 1.2 38,244 2,180 (5.7) 1,888 (4.9) 1,488 (3.9) 51.8 (12.7) 1.50 (0.27) 5.92 (1.22) 1.31 (0.70) 339 (0.9) 1,452 (3.8) 1.3 49,546 3,270 (6.6) 4,665 (9.4) 2,077 (4.2) 49.5 (11.1) 1.53 (0.29) 5.86 (1.13) 1.57 (0.84) 416 (0.8) 2,444 (4.9) All 157,864 6,072 (3.8) 4,291 (2.7) 3,634 (2.3) 49.0 (12.1) 1.08 (0.31) 5.71 (1.16) 1.09 (0.62) 917 (0.6) 3,565 (2.3)

184,261 8,463 11,216 4,670 47.4 1.09 5.74 1.43 1,049 6,472

(4.6) (6.1) (2.5) (10.9) (0.35) (1.09) (0.80) (0.6) (3.5)

a Distribution into quartiles could not be done exactly since Hp was measured with 1 decimal only. Hp haptoglobin; SD standard deviation; AMI acute myocardial infarction; HF heart failure; DM  diabetes mellitus; Q quartile.

for hypertension. The occurrences of AMI and HF were about tripled in the same range of Hp, whereas all strokes were about doubled. In men and women separately increasing levels of Hp were associated with increasing risk of AMI, IS, all strokes, and HF, whether adjusted for age alone or additionally by TC, TG, hospital-diagnosed hypertension, or DM (Tables IIIV). The risk was about 40%60% higher in quartile 4 as compared to quartile 1 with regard to IS, but 70%90% higher for AMI. For HS there was a 20% increase in risk in quartile 4 for men, but that was not seen in women. Also, the HS risk increase was seen in the younger subjects only. The HF risk gradient was clearly stronger in men than in women, but differences between genders were minor for the other end-points. The complete set of adjustment factors weakened the Hp risk gradients by less than 25%. Hazard ratios per SD increase in Hp were calculated by type of endpoint, gender, and age, with and without adjustments for TC, TG, hospital-diagnosed hypertension, and DM (Table V). Again, men and women did not vary much in their associations to AMI, but the relationships displayed by deciles of Hp were approximately linear across the whole distribution of Hp (Figure 1). The relationships were significantly stronger in men than in women for all strokes and HF (P B 0.01), but only HF showed similar progressive rise in risk by deciles of Hp as found for AMI (data not shown).

For HS men had a slight tendency towards increasing risk with increasing Hp, while in women an opposite trend was observed. Hp seemed to be a positive risk factor for all end-points in those below 65 years of age. In the elderly, associations were much weaker when measured in relative terms and slightly significantly positive for AMI and HF only. To investigate if AMI occurring prior to HF would influence the prediction of HF by Hp, a special analysis was done using the occurrence of AMI prior to a HF minus 90 days as an additional time-dependent covariate in the Cox model. The change in hazard ratio (HR) of HF for 1 SD change in Hp was practically unchanged for each gender, although this covariate itself was highly significant pin-pointing about a 15-fold increase in risk of HF if preceded by an AMI in this younger and healthy population. Comparison with TC Even if Hp was found to be a risk factor for all endpoints in this study, it should be compared with respect to strength with other conventional risk factors. As a comparator we have chosen TC since it has a classical role in coronary heart disease (CHD) epidemiology. In Figure 2 TC and Hp have been cross-classified into 4 by 4 based on quartiles for both factors. Adjusted for age, gender,

526

I. Holme et al.

Table II. Hazard ratio of ischaemic, haemorrhagic, and total stroke by quartiles of Hp per gender. Adjusted for age; and adjusted for age, TC, TG, hypertension, and DM. Haptoglobin quartiles
a

(g/L)

Ischaemic stroke HR (95% CI) 1 1.09 (1.001.19) 1.33 (1.231.45) 1.58 (1.471.71)

Haemorrhagic stroke HR (95% CI) 1 0.97 (0.821.15) 1.15 (0.981.35) 1.20 (1.031.41) 1 0.97 (0.821.15) 1.14 (0.971.34) 1.18 (1.011.38) 1 0.89 (0.711.10) 1.04 (0.851.28) 0.90 (0.731.11) 1 0.88 (0.711.09) 1.02 (0.831.25) 0.86 (0.701.06)

Total stroke HR (95% CI) 1 1.03 (0.961.11) 1.25 (1.171.34) 1.49 (1.401.59) 1 1.01 (0.941.09) 1.20 (1.121.28) 1.40 (1.311.49) 1 1.18 (1.081.28) 1.31 (1.211.43) 1.46 (1.351.58) 1 1.16 (1.071.27) 1.26 (1.161.37) 1.34 (1.241.46)

Men, adjusted for age: 5Q1 (50.8) Q1Q2 (0.81.0) Q2Q3 (1.01.3) Q3 (1.3)

Men, adjusted for age, TC, TG, hypertension and DM: 5Q1 (50.8) 1 Q1Q2 (0.81.0) 1.07 (0.981.17) Q2Q3 (1.01.3) 1.26 (1.161.37) Q3 (1.3) 1.46 (1.351.58) Women, adjusted for age: BQ1 (50.9) Q1Q2 (0.91.0) Q2Q3 (1.01.2) Q3 (1.2) 1 1.24 (1.121.38) 1.39 (1.251.53) 1.56 (1.421.72)

Women, adjusted for age, TC, TG, hypertension and DM: BQ1 (50.9) 1 Q1Q2 (0.91.00) 1.22 (1.101.35) Q2Q3 (1.001.20) 1.32 (1.201.46) Q3 (1.20) 1.43 (1.301.57)

a Distribution into quartiles could not be done exactly since Hp was measured with 1 decimal only. Hp haptoglobin; HR hazard ratio; TC total cholesterol; TG triglycerides; CI confidence interval, DM diabetes mellitus; Q quartile.

TG, hospital-diagnosed hypertension, and DM, hazard ratios for AMI are displayed using the combined lowest quartile as reference category. Both factors gave independent but TC somewhat stronger contributions than Hp to increased risk by their increasing level. There was a statistically significant (P 0.006) but numerically moderate interaction between TC and Hp on AMI hazard ratio adjusted for the covariates. The RERI index
Table III. Hazard ratio of AMI by gender specic quartilesa of Hp. Adjusted for age; and adjusted for age, TC, TG, hypertension, and DM. Men Haptoglobin quartiles g/L Adjusted for age: 5Q1 Q1Q2 Q2Q3 Q3 HR (95% CI) 1 1.16 (1.091.24) 1.45 (1.361.55) 1.92 (1.812.03) Women HR (95% CI) 1 1.11 (0.991.24) 1.35 (1.221.50) 1.99 (1.812.20)

added a significant, but numerically moderate and positive contribution to the main effects (HR 1.52 for Hp and HR 1.76 for TC), RERI 0.21 (95% CI 0.100.32). The added AP was 8.3% (95% CI 4.012.7) and S was 1.16 (95% CI 1.071.27). Similar calculations for TG revealed a highly significant (P B 0.001) interaction between TG and Hp on AMI risk, but none of the three indexes reached a significant contribution in this logistic
Table IV. Hazard ratio of HF by gender specic quartiles a of Hp. Adjusted for age; and adjusted for age, TC, TG, hypertension and DM. Men Haptoglobin quartiles (g/L) Adjusted for age: 5Q1 Q1Q2 Q2Q3 Q3 HR (95% CI) Women HR (95% CI)

1 1 1.16 (1.051.29) 0.95 (0.851.07) 1.25 (1.131.38) 1.10 (0.991.23) 1.86 (1.702.03) 1.38 (1.251.53)

Adjusted for age, TC, TG, hypertension, and DM: 5Q1 1 1 Q1Q2 1.12 (1.051.20) 1.08 (0.971.20) Q2Q3 1.33 (1.251.42) 1.25 (1.121.38) Q3 1.70 (1.601.80) 1.72 (1.561.90) Distribution into quartiles (as in Table II) could not be done exactly since Hp was measured with 1 decimal only. AMI acute myocardial infarction; HR hazard ratio; Hp haptoglobin; TC total cholesterol; TG triglycerides; CI confidence interval, DM diabetes mellitus; Q quartile.
a

Adjusted for age, TC, TG, hypertension, and DM: 5Q1 1 1 Q1Q2 1.14 (1.031.26) 0.95 (0.851.06) Q2Q3 1.19 (1.081.32) 1.06 (0.951.19) Q3 1.71 (1.561.88) 1.28 (1.161.42)
a

Distribution into quartiles (as in Table II) could not be done exactly since Hp was measured with 1 decimal only. HF heart failure; Hp haptoglobin; HR hazard ratio; TC  total cholesterol; TG triglycerides; CI confidence interval, DM diabetes mellitus.

Haptoglobin and risk of AMI, stroke, and HF

527

Table V. Comparisons of AMI, ischaemic, haemorrhagic and all stroke and HF HRs for 1 SD difference in Hp by age and gender, adjusted for age, and adjusted for age, TC, TG, hypertension and DM. AMI HR (95% CI) Adjusted for age: Age (adjusted for gender): 565 1.25 (1.241.27) 65 1.07 (1.041.10) Gender: Men 1.20 (1.191.22) Women 1.21 (1.181.23) All stroke HR (95% CI) Ischaemic stroke HR (95% CI) Haemorrhagic stroke HR (95% CI)

HF HR (95% CI)

1.17 (1.151.19) 1.01 (0.981.04) 1.14 (1.121.16) 1.09 (1.071.11)

1.18 (1.161.21) 1.00 (0.971.04) 1.14 (1.121.17) 1.10 (1.071.13)

1.06 (1.021.11) 0.94 (0.861.03) 1.07 (1.021.12) 0.96 (0.901.02)

1.27 (1.241.29) 1.05 (1.021.08) 1.20 (1.171.22) 1.10 (1.071.13)

Adjusted for age, TC, TG, hypertension and DM: Age (adjusted for gender): 565 1.22 (1.201.24) 1.15 (1.131.17) 65 1.06 (1.021.09) 1.00 (0.971.03) Gender: Men 1.18 (1.161.20) 1.12 (1.101.14) Women 1.18 (1.151.20) 1.07 (1.041.09)

1.15 (1.131.18) 0.99 (0.961.03) 1.12 (1.101.14) 1.07 (1.051.10)

1.05 (1.001.09) 0.94 (0.861.03) 1.06 (1.011.11) 0.94 (0.881.01)

1.23 (1.201.26) 1.04 (1.001.07) 1.17 (1.151.20) 1.07 (1.051.10)

AMI acute myocardial infarction; HF heart failure; Hp haptoglobin; HR hazard ratio; SD standard deviation; CI confidence interval; TC total cholesterol; TG triglycerides; DM diabetes mellitus.

model with dichotomous exposures of Hp and TC. The AMI risk spanned a 4.2-fold increase from the reference to the top combined TC-Hp category. Similar results are displayed in Figure 3 for IS (2fold increase), showing about equal predictive ability of Hp and TC. The interactions between TC and Hp were not significant (P 0.065), but were so for TG (P 0.008), but none of the three indexes were statistically significant in the logistic model. Figure 4 for HF shows Hp as the most important risk factor with a 1.5-fold increase from the lowest to the highest quartile. TC was not associated to HF when Hp was adjusted for in these analyses. There were no significant interactions for TC (P 0.080) and TG (P 0.064) and Hp on HF and no significant synergistic effects by a logistic model. Subgroup analysis In a subgroup of 85,774 men and women, where also apolipoprotein B and apolipoprotein A-1 were recorded in addition to Hp, the risk factor apoB/ apoA-1, the strongest lipid risk factor of AMI found in AMORIS (913), was compared to TC in a similar manner as between Hp and TC in Figure 2. Based on 4,254 AMI cases the risk of AMI was about 4.5 times higher in the upper joint quartile of apoB/apoA-1 as compared to the lower, whereas this relative risk for Hp was about 4.1. For IS similar numbers were 3.0 for apoB/apoA-1 and 3.1 for Hp, and for HF the numbers were 2.4 and 2.7, respectively. Thus, the predictive power of major cardiovascular events measured by relative risks may seem to be about as high for Hp as for apoB/apoA-1. However, the attributable risk for apoB/apoA-1 is

higher since more subjects were classified into the top joint quartile of TC and apoB/apoA-1 (12.8%) than that of TC and Hp (8.8%) and into the lower joint quartiles (12.1%) and (6.4%), respectively. In this subgroup there was no detectable interaction between apoB/apoA-1 and Hp on any of the three end-points (all P 0.20). Since levels of high-density lipoprotein cholesterol (HDL-C) and apoA-1 showed a slightly negative association to quartiles of Hp, analyses were repeated adjusting additionally for either HDL-C or apoA-1. Regression slopes were reduced by 10% 15%, so relationships were for all practical purposes not affected by this extra adjustment for any endpoint. Diabetes is in addition to a potential confounder also a potential effect modifier of Hp relationships to cardiovascular disease (CVD) end-points. We therefore tested for interaction between DM and Hp for each end-point by gender and age group ( B65 versus ]65 years). Generally, significant interactions were found in both genders but more so in younger than in older subjects. Relationships between Hp and end-points were clearly less evident in the DM subgroup than in non-DM subjects. For AMI and HF, but not for all stroke, only men showed a slight but positively significant association to Hp.

Discussion In this study we found that Hp, one among many acute phase reactant ISPs, is a relatively strong risk factor of AMI, stroke, and HF. There was a border-line association between Hp and HS in men and in younger subjects. The associations with AMI

528

I. Holme et al.
2.0 Gender F M Adjusted for age, TC, TG, 1.6 hospital hypertension and diabetes. 1.4

1.8

Risk ratio

1.2

1.0

<0.7 #AMIF: #AMIM: 259 838

0.70.8 0.80.9 0.91.0 1.01.1 1.11.2 1.21.3 1.31.4 1.4 1.5

>1.5 701 1915

Haptoglobin
227 704 375 955 468 1202 569 1465 505 1281 485 1104 372 993 330 759

Figure 1. Hazard ratios of acute myocardial infarction (AMI) by deciles of haptoglobin and gender adjusted for age, total cholesterol (TC), triglycerides (TG), hospital-recorded hypertension, and diabetes. Reference point not included as basis for t of regression lines.

and HF were steeper than for IS in both genders. The relationships between Hp and end-points measured in relative risk terms were clearly stronger in younger than in elderly subjects. Due to its large number of clinical events AMORIS was able to detect significant relationships with HR down to 1.05 per SD change in men and women separately for all end-points. This is down to the limit of clinical and public health interest. Even small or moderately

sized interaction effects between two risk factors could be detected in this study. The quartile analysis indicated a 70%90% increase in risk of AMI and HF from quartile 1 to 4 for men, but this gradient was only 40%60% for IS. These results were similar to those found by Engstro m et al. (4) who reported on such associations for each of five different ISPs and a combined score of those in a much smaller Swedish cohort. The

Figure 2. Hazard ratios of acute myocardial infarction (AMI) by cross-classication of quartiles of haptoglobin and total cholesterol with lowest joint quartile as reference category, adjusted for age, gender, triglycerides, hospital-recorded hypertension, and diabetes.

Figure 3. Hazard ratios of acute ischaemic stroke (IS) by crossclassication of quartiles of haptoglobin and total cholesterol with lowest joint quartile as reference category, adjusted for age, gender, triglycerides, hospital-recorded hypertension, and diabetes.

Haptoglobin and risk of AMI, stroke, and HF

529

Figure 4. Hazard ratios of heart failure (HF) by cross-classication of quartiles of haptoglobin and total cholesterol with lowest joint quartile as reference category, adjusted for age, gender, triglycerides, hospital-recorded hypertension, and diabetes.

combination of several intercorrelated ISPs into a single score can be done in many ways, and it is likely that such a score will be a better predictor of cardiovascular clinical end-points than each of its components. The method used by the Malmo group counted the number of ISPs in the upper quartile of the distribution, and they reduced this score to a dichotomous variable when testing for additive effects of ISPs over and above that of TC (35). However, by this data reduction a portion of the prediction power inherent in combining intercorrelated ISPs may get lost. For instance it is not evident from the calculations actually reported that TC and ISPs work interactively with regard to risk of AMI and HF, but possibly so for IS (4). It is known from the Belgian Interuniversity Research on Nutrition and Health (BIRNH) study that the Hp genotype 1-1 is associated to CVD death (28). Since Hp concentration is higher in genotype 1-1 than in genotypes 1-2 and 2-2 individuals, Hp 11 genotype may be a major contributor to our findings. However, since Hp genotyping is not possible in AMORIS, we cannot verify this. In AMORIS we did detect a significant interaction between TC and Hp on AMI, but not on IS or HF. However, the synergistic effect had only a small to moderate size, adding less than 15% to the AMI hazard ratios compared with each of Hp and TC considered alone. Since our logistic model was rather crude with dichotomous groupings of Hp and TC, it is conceivable that lipid-lowering treatment in those with high TC levels could be more efficient in those who also have added high Hp levels than in those with

low levels. These somewhat disparate findings as compared to those of Engstro m et al. (4) could be due to several reasons, such as longer follow-up time in their study (27 years) combined with the fact that elevated ISPs may induce larger increases in TC over time (29), the ISP score may pick up interactive associations to other ISPs than Hp, more complete set of adjustment factors accounted for in their statistical analysis, and that the AMORIS population was selected by health check-ups versus screening invitation in the Engstro m study. The tests of interaction showed that diabetics had quite different associations between Hp and end-points than non-diabetics. Unfortunately, since Hp genotyping was not possible we could not verify the suggestion that Hp genotype 2-2 is the dominant part of the excess risk in type 2 diabetics (30). We had access to only one of the five ISPs that Engstro m et al. had measured. However, by combining TC with only one ISP in a categorical analysis, AMI risk increased more than 4-fold from the lowest to the highest quartile combination. This clearly indicates that if one ISP is measured over and above TC, the ability to predict AMI may be enhanced markedly. For prediction of HF TC was not found to be significant when Hp was adjusted for. In a previous AMORIS study it was found that the ratio of apolipoprotein B (indicating the number of all potentially atherogenic and cholesterol-containing lipoprotein particles) to apolipoprotein A-1 (reflecting anti-atherogenic high-density lipoprotein cholesterol particles) predicted almost a 6-fold increase in AMI mortality from the lowest to the highest quartile combination with TC (31). In our data including also non-fatal AMI events a 4-fold increase was found when combining quartiles of TC and Hp, i.e. a considerably enhanced prediction ability of Hp when TC and the other factors were adjusted for. For IS mortality they found a 2.2-fold increase (11) in line with our finding of a 2-fold increase on IS incidence with TC and Hp quartile combinations. Since biomarkers such as apoB and apoA-1 are regarded as strong predictors of AMI and IS and clearly superior to TC as well as other conventional lipids or lipid ratios in AMORIS (9), it is noteworthy that a combined evaluation of a classic risk factor such as TC with only one ISP may give quite good predictions of AMI and IS in healthy middle-aged subjects. In AMORIS we made a special analysis in a subgroup of subjects where also apoB and apoA-1 were measured in addition to the variables in this study. By creating a similar cross-classification between the apoB/apoA-1 ratio into quartiles against TC in quartiles, keeping the adjustment factors

530

I. Holme et al. Strengths and limitations The major strength of this analysis is the large number of major cardiovascular end-points in AMORIS. Hazard ratios per 1 SD difference close to 1.05 could within important subgroups be significantly differentiated from 1.0. Thus, minimally clinical significant effect sizes could to a large degree be detected. Age and gender risk gradient contrasts had in many cases non-overlapping confidence intervals despite rather small differences. The fact that the AMORIS population was selected from health check-ups in nonhospitalized subjects needs a comment. Compared to the general population of Stockholm the all-cause mortality of the AMORIS population was about 14% lower during the study period, taking age, gender, and calendar year into account (36). Therefore, it is unlikely that this healthy cohort effect had any major impact on the results of this study. The AMORIS cohort has been shown to be similar to the general working population of Stockholm in terms of socioeconomic group and ethnicity. However, a major limitation of this study is that it did not have access to other important risk factors such as smoking, and especially use of anti-hypertensive medication etc., so adjustments for such factors would probably modify some relationships. The quality in AMI and stroke classifications rests with the quality of the discharge diagnoses and causes of death of national Swedish registers. Nevertheless, the AMI diagnosis in Swedish routine hospital care has been repeatedly evaluated and shown to be of good quality for epidemiological purposes (20). However, the quality of routine stroke diagnoses in Sweden has been less extensively evaluated and with more mixed results (1420). For HF it is known that the specificity is high but the sensitivity is low, so that an under-reporting of probably lighter cases may have occurred in our data set (21). The prevalence of hospital hypertension and DM observed by our method under-estimated its true values since hospitalizations for any cardiovascular disease, hypertension, or diabetes prior to blood sampling would occur rarely and mostly in elderly subjects. Conclusions Hp added valuable predictive information of major cardiovascular diseases over and above adjustment factors such as hospital-recorded hypertension and diabetes, TC, and TG. However, only a small-sized interaction pointing to a positive synergistic effect between TC and Hp was detected for AMI in the full population. The risk increase for increasing levels of Hp was progressive for AMI and HF, but not so for IS. The added information value over and

constant, we observed about a 4.5-fold increase in risk of fatal and non-fatal AMI from the lower to the higher combination of quartiles in this subgroup. This was just slightly higher than that found when combining quartiles of TC and Hp in the same subgroup. For IS and HF, Hp was at least as predictive as apoB/apoA-1 in these comparisons, which could indicate that the predictive power of major cardiovascular end-points over and above TC and the other factors was similar for Hp as for apoB/ apoA-1 measured in relative risk terms. However, the attributable risks were clearly lower since more subjects were classified into the lower and upper joint quartiles of apoB/apoA-1 and Hp. The reason was that the correlation between Hp and TC was smaller than that between apoB/apoA-1 and TC. It has been reported that elevated Hp may inhibit lecithin-cholesterol acyl-transferase (LCAT) activity since it will compete with it for binding to apoA-1 (32). We therefore examined whether HDL-C or apoA-1 were related to Hp quartiles. We found a significant but not strong negative association showing lower HDL-C and apoA-1 levels in quartile 4 as compared to quartile 1. However, adjusting the relationships between Hp and CVD outcomes for either HDL-C or apoA-1 did not noteworthily alter the relationships as compared to no such adjustments. The practical utility of ISPs in everyday clinical practice is not at hand as of now since they are specialized biomarkers not measured routinely. Most often hsCRP is used as inflammation marker. In AMORIS we did not measure hsCRP but only CRP according to an older established immunoturbidimetric method that did not accurately quantitate values less than 10 mg/L (33). Thus, we were not able to compare hsCRP with haptoglobin with respect to prediction ability. However, whether hsCRP is as predictive as apoB/apoA-1 of cardiovascular disease is unclear (34,35). Whether Hp is the best single age-adjusted predictor of CVD as compared to the other four ISPs used by Engstro m et al. (4) is not evident from the single-factor associations they presented. Rather, Hp seemed to be the least predictive of cardiac events in that study but as predictive of stroke. Neither can we define which of the three different subtypes (isoforms) of Hp has the strongest predictive capacity, something that in the future may be of relevance to clarify. Future studies should address which single or combination of several ISPs will give clinically useful information over and above lipoprotein components, such as non-high-density lipoprotein cholesterol or apoB/apoA-1, that are more frequently available in clinical chemical laboratories.

Haptoglobin and risk of AMI, stroke, and HF above adjustment factors was about as large as that of TC considering all events collectively, but subgroup analysis revealed that the added information was not as large as that of apoB/apoA-1. The potential for gaining better predictions of major cardiovascular end-points using ISPs seems promising, but more information is needed where lipoprotein components and more ISPs are included in the same risk models. Acknowledgements The study was supported by grants from the Gunnar and Ingmar Jungner Foundation for Laboratory Medicine, Stockholm. Declaration of interest: The authors report no conicts of interest. The authors alone are responsible for the content and writing of the paper.

531

13.

14.

15.

16.

17.

18.

References
1. Ross P. Atherosclerosis: an inammatory disease. N Engl J Med. 1999;340:11526. 2. Libby P, Ridker PM, Maseri A. Inammation and atherosclerosis. Circulation. 2002;105:113543. 3. Lind P, Engstro m G, Stavenow L, Janzon L, Lindga rde F, Hedblad B. Risk of myocardial infarction and stroke in smokers is related to plasma levels of inammation-sensitive proteins. Arterioscler Thromb Vasc Biol. 2004;24:57782. 4. Engstro m G, Lind P, Hedblad B, Stavenow L, Janzon L, Lindga rde F. Effects of cholesterol and inammationsensitive plasma proteins on incidence of myocardial infarction and stroke in men. Circulation. 2002;105:26327. 5. Engstro n P, m G, Stavenow L, Hedblad B, Lind P, Tyde Janzon L, et al. Inammation-sensitive plasma proteins and incidence of myocardial infarction in men with low cardiovascular risk. Arterioscler Thromb Vasc Biol. 2003;23: 224751. 6. Rifai N, Ridker PM. High-sensitivity C-reactive protein: a novel and promising marker of coronary heart disease. Clin Chem. 2001;47:40311. 7. Ridker PM. C-reactive protein: eighty years from discovery to emergence as a major risk marker for cardiovascular disease. Clin Chem. 2009;55:20915. 8. Quaye IK. Haptoglobin, inammation and disease. Review Trans R Soc Trop Med Hyg. 2008;102:73542. 9. Walldius G, Jungner I, Holme I, Aastveit AH, Kolar W, Steiner E. High apolipoprotein B, low apolipoprotein A-1, and improvement in the prediction of fatal myocardial infarction (AMORIS study): a prospective study. Lancet. 2001;358:202633. 10. Holme I, Aastveit AH, Jungner I, Walldius G. Relationships between lipoprotein components and risk of myocardial infarction: age, gender and short versus longer follow-up periods in the Apolipoprotein MOrtality RISk study (AMORIS). J Int Med. 2008;264:308. 11. Walldius G, Aastveit AH, Jungner I. Stroke mortality and the apoB/apoA-1 ratio: results of the AMORIS prospective study. J Int Med. 2006;259:25966. 12. Holme I, Aastveit AH, Hammar N, Jungner I, Walldius G. Relationships between lipoprotein components and risk of 19.

20.

21.

22.

23.

24.

25. 26.

27. 28.

29.

ischaemic and haemorrhagic stroke in the Apolipoprotein MOrtality RISk study (AMORIS). J Int Med. 2008;265: 27587. Walldius G, Jungner I, Kolar W, Holme I, Steiner E. High cholesterol and triglyceride values in Swedish males and females: increased risk of fatal myocardial infarction. First report from the AMORIS (Apolipoprotein related MOrtality RISk) study. Blood Pressure. 1992;1 Suppl 4:3542. Medin J, Nordlund A, Ekberg K. Increasing stroke incidence in Sweden between 1989 and 2000 among persons aged 30 to 65 years: evidence from the Swedish Hospital Discharge Register. Stroke. 2004;35:104751. Lindblad U, Ra stam L, Ranstam J, Peterson M. Validity of register data on acute myocardial infarction and acute stroke: the Skaraborg Hypertension Project. Scand J Soc Med. 1993;21:39. Stegmayr B, Asplund K. Measuring stroke in the population: quality of routine statistics in comparison with a populationbased stroke registry. Neuroepidemiology. 1992;11:20413. Stegmayr B, Glader E-L. Five years with Riks-Stroke. The Swedish national register for quality assessment of acute stroke care. Available at: www.riks-stroke.org (assessed on 27 July 2009). Glader EL, Stegmayr B, Norrving B, Tere nt A, Hulter sberg K, Wester PO, et al. Riks-Stroke Collaboration. Sex A differences in management and outcome after stroke: a Swedish national perspective. Stroke. 2003;34:19705. Appelros P, Nydevik I, Seiger A, Tere nt A. High incidence rebro, Sweden: Further support for rates of stroke in O regional incidence differences within Scandinavia. Cerebrovasc Dis. 2002;14:1618. Hammar N, Alfredsson L, Rose n M, Spetz C-L, Kahan T, Ysberg A-S. A national record linkage to study acute myocardial infarction incidence and case fatality in Sweden. Int J Epidemiol. 2001;30 Suppl 1:S304. Ku mler T, Gislason GH, Kirk V, Bay M, Nielsen OW, Kber L, et al. Accuracy of a heart failure diagnosis in administrative registers. Eur J Heart Fail. 2008;10:65860. rnlo Ingelsson E, .A w J, Sundstro m J, Lind L. The validity of a diagnosis of heart failure in a hospital discharge register. Eur J Heart Fail. 2005;7:78791. Jungner I, Walldius G, Holme I, Kolar W, Steiner E. Apolipoprotein B and A-1 in relation to serum cholesterol and triglycerides in 43 000 Swedish males and females. Int J Clin Lab Res. 1992;21:24755. Jungner I, Marcovina SM, Walldius G, Holme I, Kolar W, Steiner E. Apolipoprotein B and A-1 values in 147 576 Swedish males and females, standardized according to the World Health Organization-International Federation of Clinical Chemistry First International Reference Materials. Clin Chem. 1998;44:16419. Hosmer DW, Lemeshow S. Condence interval estimation of interaction. Epidemiology. 1992;3:4526. Lundberg M, Fredlund P, Hallquist J, Dederichsen F. A SAS program calculating three measures of interaction with condence intervals. Epidemiology. 1996;7:6556. Rothmann KJ. The estimation of synergism and antagonism. Am J Epidemiol. 1976;103:50611. De Bacquer D, De Backer G, Langlois M, Delanghe J, Kesteloot H, Kornitzer M. Haptoglobin polymorphism as a risk factor for coronary heart disease mortality. Atherosclerosis. 2001;157:1616. Engstro m G, Hedblad B, Janzon L, Lindga rde F. Long-term change in cholesterol in relation to inammation-sensitive plasma proteins: A longitudinal study. Ann Epidemiol. 2007;17:5763.

532

I. Holme et al.
lipid measures, lipid ratios, and CRP as risk factors for cardiovascular disease in women. JAMA. 2005;294:32633. 35. Sierra-Johnson J, Fisher RM, Romero-Corral A, Somers VK, hrvik J, et al. Concentration of apolipoLopez-Jimenez F, .O protein B is comparable with the apolipoprotein B/ apolipoprotein A-1 ratio and better than routine clinical lipid measurements in predicting coronary heart disease mortality: ndings from a multi-ethnic US population. Eur Heart J. 2009;30:7107. 36. Holzmann MJ, Jungner I, Walldius G, Ivert T, Nordqvist T, stergren J, et al. Apolipoproteins B and A-I, standard lipid O measures and incidence of myocardial infarction in men and women, with or without chronic kidney disease. Study IV in Thesis for doctorial degree (Ph.D.). In: Holzmann M. Renal insufciency, mortality and myocardial infarction. Karolinska Institutet, Stockholm; 2008.

30. Levy AP. Modication of HDL structure and function in individuals with diabetes and the haptoglobin 2-2 genotype. Clin Lipidology. 2009;4:58. 31. Walldius G, Jungner I. Apolipoprotein B and apolipoprotein A-1: risk indicators of coronary heart disease and targets for lipid-modifying therapy. J Int Med. 2004;255:188205. 32. Salvatore A, Cigliano L, Bucci EM, Corpillo D, Velasco S, Carlucci A, et al. Haptoglobin binding to apolipoprotein A-1 prevents damage from hydroxyl radicals on its stimulatory activity of the enzyme lecithin-cholesterol acyl-transferase. Biochemistry. 2007;46:1115868. 33. Orion Diagnostics, Helsinki, Finland. Rapid immunochemical determination of C-Reactive Protein using clinical chemistry analyzers (January 1982) www.oriondiagnostica. com (assessed on 27 July 2009). 34. Ridker PM, Rifai N, Cook NR, Bradwin G, Buring JE. NonHDL cholesterol, apolipoproteins A-1 and B100, standard